design of an artificial decision maker for a human-based social...
TRANSCRIPT
1
Design of an Artificial Decision Maker
for a Human-based Social Simulation
– Experience of the SimParc Project (Submitted to Cargese’09 Seminar)
Jean-Pierre Briot1,2
, Alessandro Sordoni1, Eurico Vasconcelos
2
Vinícius Sebba Patto1, Diana Adamatti
3, Marta Irving
4, Carlos Lucena
2
(1) Laboratoire d’Informatique de Paris 6, Université Pierre et Marie Curie – CNRS, Paris,
France
(2) Computer Science Department, Pontifícia Universidade Católica, Rio de Janeiro, RJ, Brazil
(3) Faculdade de Tecnologia (FTEC), Caxias do Sul, RS, Brazil
(4) EICOS Postgrad Program, Universidade Federal do Rio de Janeiro, Rio de Janeiro, RJ, Brazil
Corresponding author : [email protected]
Abstract
This paper addresses an ongoing experience in the design of an artificial agent
taking decisions in a social simulation (more precisely, a role playing game)
populated by human agents and by artificial agents. At first, we will present our
current research context, an ongoing research project aimed at computer-based
support for participatory management of protected areas (and more specifically
national parks) in order to promote biodiversity conservation and social inclusion.
Our applicative objective is to help various stakeholders (e.g., environmentalist
NGOs, communities, tourism operators, public agencies) to collectively understand
conflict dynamics for natural resources management and to explore negotiation
management strategies for the management of parks, one of the key issues linked to
biodiversity conservation. Our approach combines techniques such as: distributed
role playing games (serious games), geographical information systems, support for
negotiation between players, and insertion of various types of artificial agents
(virtual players, decision making agents, assistant agents). In this paper, we will
focus on the design of the decision making agent architecture for the park manager
agent, the rationales for his decision and how he takes into account the
preferences/votes from the players of the game and may justify/explain his decisions.
Keywords
Social simulation, serious games, participatory management, decision making, multi-
agent, negotiation, argumentation, explanation
1. Introduction
In this paper, we discuss the issue for an artificial agent to take decisions in a
social simulation. The type of social simulation that we refer to here is a social
simulation populated by humans. More precisely, we consider a role playing game
2
(“serious game”) where humans play some role and discuss, negotiate and take
decisions about a common domain, in our case environment management decisions.
We are now in the process of inserting artificial agents into the human-based
social simulation [Briot et al. 2008]. One of the ideas is to possibly replace some of
the human players by artificial players (artificial agents). The social simulation will
therefore become hybrid, with human and artificial agents. A first motivation is to
address the possible absence of sufficient number of human players for a game
session [Adamatti et al. 2007]. But this will also allow more systematic experiments
about specific configurations of players profiles, because of artificial players’
objective, deterministic and reproducible behaviors.
More precisely, we are considering three types of artificial agents: artificial park
manager, artificial players and assistant agents. In this paper we focus on the design
of the artificial park manager agent. Its objective is to take decision based on its own
analysis of the situation and on the proposals by the players. The agent is also able to
explain its decision based on its chain of argumentation.
In this paper, after introducing the SimParc project, its role playing game and its
computer support, we describe the park manager agent objectives, architecture and
implementation.
2. The SimParc Project
2.1 Project Motivation
A significant challenge involved in biodiversity management is the management of
protected areas (e.g., national parks), which usually undergo various pressures on
resources, use and access, which results in many conflicts. This makes the issue of
conflict resolution a key issue for the participatory management of protected areas.
Methodologies intending to facilitate this process are being addressed via bottom-up
approaches that emphasize the role of local actors. Examples of social actors involved
in these conflicts are: park managers, local communities at the border area, tourism
operators, public agencies and NGOs. Examples of inherent conflicts connected with
biodiversity protection in the area are: irregular occupation, inadequate tourism
exploration, water pollution, environmental degradation and illegal use of natural
resources.
Our SimParc project focuses on participatory parks management. (The origin of
the name SimParc stands in French for “Simulation Participative de Parcs”) [Briot et
al. 2007]. It is based on the observation of several case studies in Brazil. However,
we chose not to reproduce exactly a real case, in order to leave the door open for
broader game possibilities [Irving et al. 2007]. Our project aim is to help various
stakeholders at collectively understand conflicts and negotiate strategies for handling
them.
3
2.2. Approach
Our initial inspiration is the companion modeling (ComMod) approach about
participatory methods to support negotiation and decision-making for participatory
management of renewable resources [Barreteau et al. 2003]. Their pioneer method,
called MAS/RPG, consists in coupling multi-agent simulation (MAS) of the
environment resources with a role-playing game (RPG) [Barreteau 2003]. Hence,
stakeholders may understand and explore the consequences of their decisions. The
RPG acts like a “social laboratory”, because players of the game can try many
possibilities, without real consequences.
Recent works proposed further integration of role-playing into simulation, and the
insertion of artificial agents, as players or as assistants. Participatory simulation and
its incarnation, the Simulación framework [Guyot and Honiden 2006], focused on a
distributed support for role-playing and negotiation between human players. All
interactions are recorded for further analysis (thus opening the way to automated
acquisition of behavioral models) and assistant agents are provided to assist and
suggest strategies to the players. The Games and Multi-Agent-based Simulation
(GMABS) methodology focused on the integration of the game cycle with the
simulation cycle [Adamatti et al. 2007]. It also innovated in the possible replacement
of human players by artificial players.
2.3 Game Objectives
Current SimParc game has an epistemic objective: to help each participant
discover and understand the various factors, conflicts and the importance of dialogue
for a more effective management of parks. Note that this game is not (or at least not
yet) aimed at decision support (i.e., we do not expect the resulting decisions to be
directly applied to a specific park).
The game is based on a negotiation process that takes place within the park
council. This council, of a consultative nature, includes representatives of various
stakeholders (e.g., community, tourism operator, environmentalist, nongovernmental
association, water public agency…). The actual game focuses on a discussion within
the council about the “zoning” of the park, i.e. the decision about a desired level of
conservation (and therefore, use) for every sub-area (also named “landscape unit”) of
the park. We consider nine pre-defined potential levels (that we will consider as
types) of conservation/use, from more restricted to more flexible use of natural
resources, as defined by the (Brazilian) law. Examples are: Intangible, the most
conservative use, Primitive and Recuperation.
The game considers a certain number of players’ roles, each one representing a
certain stakeholder. Depending on its profile and the elements of concerns in each of
the landscape units (e.g., tourism spot, people, endangered species…), each player
will try to influence the decision about the type of conservation for each landscape
unit. It is clear that conflicts of interest will quickly emerge, leading to various
4
strategies of negotiation (e.g., coalition formation, trading mutual support for
respective objectives, etc).
A special role in the game is the park manager. He is a participant of the game, but
as an arbiter and decision maker, and not as a direct player. He observes the
negotiation taking place between players and takes the final decision about the types
of conservation for each landscape unit. His decision is based on the legal framework,
on the negotiation process between the players, and on his personal profile (e.g., more
conservationist or more open to social concerns) [Irving 2006]. He may also have to
explain his decision, on players demand. Players and the park manager roles may be
played by humans or by artificial agents.
2.4 Game Cycle
The game is structured along six steps, as illustrated in Figure 1. At the beginning
(step 1), each participant is associated to a role. Then, an initial scenario is presented
to each player, including the setting of the landscape units, the possible types of use
and the general objective associated to his role. Then (step 2), each player decides a
first proposal of types of use for each landscape unit, based on his/her understanding
of the objective of his/her role and on the initial setting. Once all players have done
so, each player’s proposal is made public. In step 3, players start to interact and to
negotiate on their proposals. This step is, in our opinion, the most important one,
where players collectively build their knowledge by means of an argumentation
process. In step 4, they revise their proposals and commit themselves to a final
proposal for each landscape unit. In step 5, the park manager makes the final
decision, considering the negotiation process, the final proposals and also his
personal profile (e.g., more conservationist or more sensitive to social issues). Each
player can then consult various indicators of his/her performance (e.g., closeness to
his initial objective, degree of consensus, etc.). He can also ask for an explanation
about the park manager decision rationales. The last step (step 6) “closes” the
epistemic cycle by considering the possible effects of the decision. In the current
game, the players provide a simple feedback on the decision by indicating their level
of acceptance of the decision. A new negotiation cycle may then start, thus creating a
kind of learning cycle. The main objectives are indeed for participants: to understand
the various factors and perspectives involved and how they are interrelated; to
negotiate; to try to reach a group consensus; and to understand cause-effect relations
based on the decisions.
5
Figure 1. The six steps of SimParc game.
2.5 Towards Evaluating the Viability of Decisions
As described in previous section, the last step of the game “closes” the cycle by
considering the possible effects of the decision. In the current game, players provide a
simple feedback on the decision by selecting their level of acceptance of the decision.
In a future project, we would like to introduce some technical evaluation of the
quality and viability of the decision (e.g., considering the survival of an endangered
species). Therefore, we plan to identify cases of usage conflicts (e.g., between
tourism and conservation of an endemic species) and model the dynamics of the
system (in an individual-based/multi-agent model or/and in an aggregated model).
We would then like to explore the use of viability theory to evaluate the viability of
the decision. Note that in our project current stage, we are concerned with credibility
and not yet with realism because our objective is epistemic and not about producing
an (hypothetical) optimal decision.
3. Computer Support for Role Playing Games
Our current prototype benefited from our previous experiences (game sessions and
a first prototype) and has been based on a detailed design process. Based on the
system requirements, we adopted Web-based technologies (more precisely J2E and
JSF) that support the distributed and interactive character of the game as well as an
easy deployment. Figure 2 shows the general architecture and communication
structure of SimParc prototype version 2. In this second prototype, distributed users
(the players and the park manager) interact with the system mediated internally by
communication broker agents (CBA). The function of a CBA is to abstract the fact
that each role may be played by a human or by an artificial agent. A CBA also
translates user messages in http format into multi-agent KQML format and vice
6
versa. For each human player, there is also an assistant agent offering assistance
during the game session.
During the negotiation phase (see in Figure 3 an example of the corresponding
user interface [Vasconcelos et al. 2008]), players (human or artificial) negotiate
among themselves to try to reach an agreement about the type of use for each
landscape unit (sub-area) of the park. A Geographical Information System (GIS)
offers to users different layers of information (such as flora, fauna and land
characteristics) about the park geographical area. All the information exchanged
during negotiation phase namely users’ logs, game configurations, game results and
general management information are recorded and read from a PostgreSql database.
The game and the supporting prototype have been tested through two game
sessions by expert players (including a professional park manager) in January 2009
(see Figure 4).
Figure 2. SimParc version 2 general architecture
7
Figure 3. Example of SimParc negotiation graphical user interface
Figure 4. SimParc session (January 2009)
8
4. Inserting Artificial Agents in the SimParc Game
We are currently inserting artificial agents into the prototype. We consider three
types of artificial agents: the park manager, the artificial players and the assistants.
The park manager acts as an arbitrator in the game, making a final decision for
types of conservation for each landscape unit and explains its decision to all players.
He may be played by a human or by an artificial agent. We have implemented a
prototype of an artificial park manager, based on 2 steps: (1) internal/individual
decision by the park manager, based on some argumentation model; (2) merging of
the decision by the manager with the votes by the players, based on decision theory
(social choice). Traces of argumentation may be used for explaining the rationale of
the decision. The artificial park manager architecture is detailed in next section.
Concerning artificial players, we refer to previous experience about virtual players
in the ViP-JogoMan system [Adamatti et al., 2007]. The idea is to possibly replace
some of the human players by artificial players (artificial agents). The two main
motivations are: (1) the possible absence of sufficient number of human players for a
game session and (2) the need for testing in a systematic way specific configurations
of players’ profiles. The artificial players will be developed along park manager
existing architecture. We plan to use its argumentation capabilities to generate the
negotiation process. In parallel, we also explore using automated analysis of recorded
traces of interaction between human players in order to infer models of artificial
players. In some previous work [Guyot and Honiden 2006], genetic programming had
been used as a technique to infer interaction models, but we will also intend to use
alternative induction and machine learning techniques, e.g., inductive logic
programming.
The last type of artificial agent is an intelligent assistant agent. This agent is
designed to assist a player by performing two main tasks: (1) to help participants in
playing the game, e.g.: the assistant agent tells the player when he should make
decisions; what are the phases of the game; what should be done in each phase; etc.;
(2) to help participants during the negotiations. For this second task, we would like to
avoid intrusive support, which may interfere in his decision making cognitive
process. We have selected some objectives, e.g., to identify other players' roles with
similar or dissimilar goals, which may help the human player to find possible
coalitions or conflicts. An initial prototype implementation of an assistant agent has
been implemented.
Further details about SimParc artificial players and assistant agents may be found
in [Briot et al. 2008] and in future publications. Some advanced interface has also
been designed for human players dialogue and negotiation support and is detailed in
[Vasconcelos et al. 2008]. Now we will detail the rationale and architecture of our
automated park manager who makes the final decision.
9
5. Park Manager Artificial Agent
In this section, we propose an agent architecture to implement park manager
cognitive decision rationale. As we summarized before, our decision model is based
on two mechanisms. These mechanisms could be viewed as modules of decision
subprocesses. We believe that complex decision making is achievable by sequential
organization of these modules. Before proceeding to the description of our agent
architecture, we present some more detailed motivation for it.
5.1 Objectives
Participatory management aims to emphasize the role of local actors in managing
protected areas. However, the park manager is the ultimate arbiter of all policy on
devolved matters. He acts like an expert who decides on validity of collective
concerted management policies. Moreover, he is not a completely fair and objective
arbiter: he still brings his personal opinions and preferences in the debate. Therefore,
we aim to develop an artificial agent modeling the following behaviors.
Personal preferential profile; park manager decision-making process is supposed to
be influenced by its sensibility to natural park stakes and conflicts. In decision theory
terms, we can affirm that park manager’s preferential profile could be intended as a
preference relation over conservation policies. One of the key issues is to understand
that we cannot define a strict bijection between preferential profile and preference
relation. Agent’s preference relation is partially dependent on natural park resources
and realities. Moreover, this relation is not likely to be an order or a preorder. Hence,
our agent must be able to generate its preference relation according with its
preferential profile. We distinguish two preferential profiles:
i. Preservationist, aims to preserve ecosystems and the natural environment.
ii. Socio-conservationist, generally accepts the notion of sustainable yield - that
man can harvest some forest or animal products from a natural environment
on a regular basis without compromising the long-health of the ecosystem.
Taking into account stakeholders’ decisions; a participatory decision-making leader
seeks to involve stakeholders in the process, rather than taking autocratic decisions.
However, the question of how much influence stakeholders are given may vary on
manager’s preferences and beliefs. Hence, our objective is to model the whole
spectrum of participation, from autocratic decisions to fully democratic ones. To do
so, we want the park manager agent to generate a preference preorder over
conservation policies. This is because it should be able to calculate the distance
between any two conservation policies. This way, we can merge stakeholders’
preference preorders with manager’s one to establish one participatory final decision.
Autocratic/democratic manager attitude will be modeled by an additional parameter
during the merge process.
10
Expert decision; the park manager’s final decision must consider legal constraints
related to environmental management; otherwise, non-viable decisions would be
presented to the players, thus invalidating game’s learning objectives. These
constraints are directly injected in the cognitive process of the agent. Hence, the agent
will determine a dynamic preference preorder over allowed levels of conservation
(according to its preferential profile).
Explaining final decision; In order to favor the learning cycle, the park manager
agent must be able to explain its final decision to the players. We can consider that
the players could eventually argue about its decision; the agent should then defend its
purposes using some kind of argumentative reasoning. Even if such cases will be
explored in future work, it is our concern to conceive a cognitive architecture which
provides a good basis for managing these situations.
5.2 Architecture Overview
Let us now present an architecture overview of the park manager agent. As
depicted in Figure 5, agent’s architecture is structured in two phases. We believe that
sequential decision-making mechanisms can model complex cognitive behaviors
along with enhanced explanation capabilities.
Figure 5. Park Manager Agent 2-steps decision process
The first decision step concerns agent’s individual decision-making process: the
agent deliberates about the types of conservation for each landscape unit. Broadly
speaking, park manager agent builds its preference preorder over allowed levels of
conservation. An argumentation-based framework is implemented to support the
decision making.
The next step of our approach consists of taking account of players’ preferences.
The result of the execution is the modified park manager decision, called agent
participatory decision, according to stakeholder’s preferences.
11
5.2.1 Agent Individual Decision
Recently, argumentation has been gaining increasing attention in the multi-agent
community. Autonomous and social agents need to deliberate under complex
preference policies, related to the environment in which they evolve. Generally,
social interactions bring new information to the agents. Hence, preference policies
need to be dynamic in order to take account of newly acquired knowledge. Dung’s
work [Dung 1995] proposes formal proof that argumentation systems can handle
epistemic reasoning under open-world assumptions, usually modeled by non-
monotonic logics. Argumentation thus becomes an established approach for
reasoning with inconsistent knowledge, based on the construction and the interaction
between arguments. Recently, some research has considered argumentation systems
capabilities to model practical reasoning, aimed at reasoning about what to do
[Hulstijn and Torre 2003, Amgoud and Kaci 2004, Rahwan and Amgoud 2006]. It
is worth noticing that argumentation can be used to select arguments that support
available desires and intentions. Consistent knowledge can generate conflicting
desires. An agent should evaluate pros and cons before pursuing any desire. Indeed,
argumentative deliberation provides a mean for choosing or discarding a desire as an
intention.
We could argue that open-world assumptions don’t hold in our context. Agent’s
knowledge base isn’t updated during execution, since it’s not directly exposed to
social interactions. Knowledge base and inference rules consistency-checking
methods are, therefore, not necessary. However, one key aspect here is to conceive an
agent capable of explaining its policy making choices; our concern is to create
favorable conditions for an effective and, thus closed, learning cycle. We believe that
argumentation “tracking” represents an effective choice for accurate explanations.
Conflicts between arguments are reported to the players, following agent’s reasoning
cycle, thus enhancing user comprehension.
From this starting position, we have developed an artificial agent on the basis of
Rahwan and Amgoud’s work [Rahwan and Amgoud 2006]. The key idea is to use
argumentation system to select the desires the agent is going to pursue: natural park
stakes and dynamics are considered in order to define objectives for which to aim.
Hence, decision-making process applies to actions, i.e. levels of conservation, which
best satisfy selected objectives. In order to deal with arguments and knowledge
representation, we use first-order logic. Various inference rules were formulated with
the objective of providing various types of reasoning capability.
For example, a simple rule for generating desires from beliefs, i.e. natural park
stakes, is:
Fire Avoid_Fires, 4
where Fire (fire danger in the park) is a belief in agent’s knowledge base and
Avoid_Fires is the desire that is generated from the belief. The value 4 represents the
12
intensity of the generated desire.
Examples of rules for selecting actions, i.e. level of conservation, from desires are:
Primitive Avoid_Fires, 0.4
Intangible Avoid_Fires, 0.8
where Primitive, Intangible represent levels of conservation and values 0.4, 0.8
represent their utility in order to satisfy the corresponding desire.
5.2.2 Agent Participatory Decision
Despite participatory ideals, a whole spectrum of park managers, from autocratic
to fully democratic ones, can be measured, depending on how more participatory and
democratic decision-making is operationalized. We propose a method, fitted into the
social-choice framework, in which participatory attitude is a model parameter.
In a real case scenario, a decision-maker would examine each stakeholder's
preferences in order to reach the compromise that best reflects its participatory
attitude. Our idea is to represent this behavior by weighting each player's vote
according to manager’s point of view.
Figure 6. Park Manager Agent Participatory Decision
This concept is illustrated in Figure 6. The process is structured in two phases.
Firstly, manager agent injects its own preferences into players' choices by means of
an influence function describing agent's participatory attitude. Stronger influence
translates into more autocratic managers. Secondly, modified players' choices are
synthesized, using an aggregation function, i.e. Condorcet voting method. The result
of the execution will be the agent participatory decision.
13
Example. Let the following be players' choices, where > is a preference relation
(a > b means “a is preferred to b”) and A={Intangible, Primitive, Extensive} the
candidates’ set. Players’ choices are converted into numeric vectors specifying the
candidates’ rank (each column corresponds to a candidate) for each vote:
player_1: Intangible > Primitive > Extensive, v1 = ( 3, 2, 1 )
player_2: Extensive > Primitive > Intangible, v2 = ( 1, 2, 3 )
player_3: Primitive > Extensive > Intangible, v3 = ( 1, 3, 2 )
Let Manager individual decision be:
manager_individual: Extensive > Primitive > Intangible, vM = ( 1, 2, 3 )
Let the following be the influence function:
Modified player’s vectors will be :
mv1 = ‹ f ( v1(1), vM(1) ), ‹ f ( v1(2), vM(2) ), ‹ f ( v1(3), vM(3) ) › = ( 1.5, 2, 0.5 )
mv2 = ( 1, 2, 3 )
mv3 = ( 1, 3, 2 )
In order to find manager participatory decision, we apply the Choquet integral Cμ
[Choquet 1953] choosing a symmetric capacity measure μ(S) = |S|2
/ |A|2, where A is
the candidates set.
Cμ(Intangible) = 1.05, Cμ(Primitive) = 2.12, Cμ(Extensive) = 1.27
The result of the execution will then be:
manager_participatory: Primitive > Extensive > Intangible.
Further details about architecture formal background and implementation are
reported in [Sordoni, 2008] and will be in future publications.
5.3 Examples of results
Presented manager agent architecture and its first implementation were tested over
different scenarios. Tests of the park manager agent were conducted offline in
laboratory and have been validated by team experts. We are currently completing the
integration of the park manager agent into the prototype. We will then soon conduct
new session tests with human players and park manager agent.
14
We report hereafter an example of explanation for manager’s decision over a
landscape unit. Let manager individual decision be the following:
Manager_individual: Intangible > Recuperation
Arguments for Intangible are:
Endangered_Species & Tropical_Forest Maximal_Protection
Intangible Maximal_Protection
Arguments for Recuperation are:
Fire & Agricolture_Activities Recover_deteriorated_zone
Recuperation Recover_deteriorated_zone
5.4 Implementation framework
The architecture presented in this paper has been implemented in Jason multi-
agent platform [Bordini and Hubner, 2007]. Besides interpreting the original
AgentSpeak(L) language, thus disposing of logic programming capabilities, Jason
also features extensibility by user-defined internal actions, written in Java. Hence, it
has been possible to easily implement aggregation methods.
6 Conclusion
In this paper, we have presented the SimParc project, a role-playing serious game
aimed at computer-based support for participatory management of protected areas.
The lack of human resources implied in RPG gaming process acts as a constraint to
the fulfillment of pedagogical and epistemic objectives. In order to guarantee an
effective learning cycle, park manager role must be played by a domain expert.
Required expertise obviously narrows game’s autonomy and limits its context of
application. Our solution to this problem is to insert artificial agents into the game. In
this paper, we focused on the architecture to implement park manager cognitive
decision rationale. It can justify its behavior and generate a participatory decision.
Our argumentation system is based on [Amgoud and Kaci 2004, Rahwan and
Amgoud 2006]: conflicts between arguments can be reported thus enhancing user
comprehension. Moreover, we presented a decision theory framework responsible for
generating a participatory decision. To the best of our knowledge and belief, this
issue has not yet been addressed in the literature. The game and the supporting
prototype have already been tested through game sessions by expert players
(including a professional park manager) in January 2009. The final integration of the
complete SimParc prototype (with artificial agents) is under completion and we will
soon start to test it. Besides the project specific objectives, we also plan to study the
possible generality of our prototype for other types of human based social
15
simulations. In the current architecture of the artificial park manager, only static
information about the park and about the votes of the players are considered. We are
considering exploring how to introduce dynamicity in the decision model, taking into
account the dynamics of negotiation among the players (the evolution of player’s
decisions during negotiation).
Acknowledgments
We thank the other current members of the project: Isabelle Alvarez, Gustavo Melo,
Altair Sancho and Davis Sansolo for their current participation; and also: Ivan
Bursztyn and Paul Guyot for their past participation. This research is funded by: the
ARCUS Program (French Ministry of Foreign Affairs, Région Ile-de-France and
Brazil) and the MCT/CNPq/CT-INFO “Grandes Desafios” Program (Project No
550865/2007-1). Some additional individual support is provided by Alßan (Europe),
CAPES and CNPq (Brazil) fellowship programs.
References
[Adamatti et al. 2007] Adamatti, D.F., Sichman, J.S. and Coelho, H. Virtual Players:
From Manual to Semi-Autonomous RPG, Proceedings of the International
Modeling and Simulation Multiconference (IMSM'07), Buenos Aires, Argentina,
The Society for Modeling & Simulation International (SCS), February 2007.
[Amgoud and Kaci 2004] Leila Amgoud and Souhila Kaci. On the generation of
bipolar goals in argumentation-based negotiation. In Iyad Rahwan, Pavlos
Moraitis, and Chris Reed, editors, Argumentation in Multi-Agent Systems:
(Proceedings of the First International Workshop (ArgMAS’04): Expanded and
Invited Contributions), volume 3366 of Lecture Notes in Computer Science.
Springer Verlag, Berlin, Germany, 2005.
[Hulstijn and Torre 2003] Joris Hulstijn and Leendert van der Torre. Combining goal
generation and planning in an argumentation framework. In Tom Heskes, Peter
Lucas, Louis Vuurpijl, and Wim Wiegerinck, editors, Proceedings of the 15th
Belgium-NetherlandsConference on Artificial Intelligence (BNAIC 2003),
Katholieke Universiteit Nijmegen, October 2003, pp. 155–162.
[Barreteau 2003] Barreteau, O. The Joint Use of Role-Playing Games and Models
Regarding Negotiation Processes: Characterization of Associations, Journal of
Artificial Societies and Social Simulation, Vol. 6, No 2, 2003.
[Barreteau et al. 2003] Barreteau, O et al. Our Companion Modelling Approach,
Journal of Artificial Societies and Social Simulation, Vol. 6, No 2, 2003.
[Bordini and Hubner 2007] Bordini, R. and Hubner, J. JASON: A Java-based
AgentSpeak Interpreter used with Saci for Multi-Agent Distribution over the Net”,
Available at http://jason.sourceforge.net/.
16
[Briot et al. 2007] Briot, J.-P., Guyot, P. and Irving, M. Participatory Simulation for
Collective Management of Protected Areas for Biodiversity Conservation and
Social Inclusion, Proceedings of the International Modeling and Simulation
Multiconference (IMSM'07), Buenos Aires, Argentina, The Society for Modeling
& Simulation International (SCS), February 2007, pp. 183–188.
[Briot et al. 2008] Briot, J.-P., Vasconcelos, E., Adamatti, D., Sebba, V., Irving, M.,
Barbosa, S., Furtado, V., Lucena, C. A Computer-based Support for Participatory
Management of Protected Areas: The SimParc Project. Proceedings of XXVIIIth
Congress of Computation Brazilian Society (CSBC'08) – Seminário Integrado de
Software e Hardware “Grandes Desafios”, Belem, PA, Brazil, July 2008, pp. 1–
15.
[Choquet 1953] Choquet, G. Theory of capacities, Journal of Fourier Institute, No 5,
1953, pp. 131-295.
[Guyot and Honiden 2006] Guyot, P. and Honiden, S. Agent-Based Participatory
Simulations: Merging Multi-Agent Systems and Role-Playing Games, Journal of
Artificial Societies and Social Simulation, Vol. 9, No 4, 2006.
[Irving 2006] Irving, M. (Ed.). Áreas Protegidas e Inclusão Social: Construindo
Novos Significados, Rio de Janeiro: Aquarius, 2006.
[Irving et al. 2007] Irving, M.A., Briot, J.-P., Bursztyn, I., Guyot, P., Melo, G.,
Sancho, A., Sansolo, D., Sebba Patto V. and Vasconcelos, E. Simparc: Computer
Supported Methodological Approach for Constructing Democratic Governance in
National Parks Management, Proceedings of IUCN II Congreso Latinoamericano
de Parques Nacionales y Otras Areas Protegidas, Bariloche, Argentina,
September–October 2007.
[Rao and Georgeff 1991] Rao, A. and Georgeff, M. Modelling Rational Agents
within a BDI-Architecture, Proceedings of the 2nd International Conference on
Principles of Knowledge Representation and Reasoning, 1991, pp. 473–484.
[Rahwan and Amgoud 2006] Iyad Rahwan and Leila Amgoud. An Argumentation-
based Approach for Practical Reasoning . In Gerhard Weiss and Peter Stone,
editors, 5th International Joint Conference on Autonomous Agents & Multi Agent
Systems, AAMAS’2006, Hakodate, Japan, New York, USA, ACM Press, 2006, pp.
347–354.
[Sordoni, 2008] Alessandro Sordoni. Master’s Degree Thesis. http://www-
poleia.lip6.fr/~sordonia/master-thesis.pdf
[Vasconcelos et al. 2008] Vasconcelos, E., Briot, J.-P., Barbosa, S., Furtado, V. and
Irving, M. A User Interface to Support Dialogue and Negotiation in Participatory
Simulations, In Proceedings of 9th International Workshop on Multi-Agent-Based
Simulation (MABS’2008), Estoril, Portugal, May 2008, pp. 47–58.